This is the mail archive of the gcc-help@gcc.gnu.org mailing list for the GCC project.

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]
Other format:	[Raw text]

Re: FAIL in movm pattern

From: Ian Lance Taylor <iant at google dot com>
To: Aurelien Buhrig <aurelien dot buhrig dot gcc at gmail dot com>
Cc: gcc-help at gcc dot gnu dot org
Date: Thu, 06 Oct 2011 11:16:26 -0700
Subject: Re: FAIL in movm pattern
References: <CAJ-J9_UK0UfK12=oq8e_1DFRwSH=nOm6hWK5hf7yw8U1O=Vhiw@mail.gmail.com>

Aurelien Buhrig <aurelien.buhrig.gcc@gmail.com> writes:

> When is it safe to generate a FAIL in the define_expand of a mov pattern?
> Is it enough to test !can_create_pseudo_p() before a FAIL ?
>
> If no such test is performed, some mov insns can be omitted (ex:
> expand_call/store_one_arg), without any error message or warning and
> it can be difficult to debug it.
> Is it a desirable behavior?

In general, a mov pattern may not FAIL.  The reload pass relies on the
ability to generate moves even when no pseudo-registers are available.

It's true that the middle-end does not use assertions to enforce this.

Ian

References:
- FAIL in movm pattern
  - From: Aurelien Buhrig

Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]